LiteLLM vulnerabilities could expose AI gateway systems

Security researchers at Check Point Research (CPR) have found two vulnerabilities in LiteLLM, an open-source gateway that lets applications connect to large language models from providers including OpenAI and Anthropic. LiteLLM supports about 140 AI providers and has more than 90 million monthly downloads on GitHub, along with more than 61,000 stars.

CPR says the flaws affect LiteLLM versions earlier than 1.83.9. The first, CVE-2026-35029, is an access-control bug rated 8.7/10, or high severity. It could let users with minimal privileges reach administrative functions, read server files, overwrite administrator credentials, and run malicious code.

The second, CVE-2026-59823, is a server-side request forgery bug rated 5.3/10, or medium severity. CPR says it could let users direct the gateway to send requests carrying AI provider authentication keys to third-party servers, potentially exposing credentials that can provide billable access to AI services.

LiteLLM released fixes for both vulnerabilities in version 1.83.9. CPR says organizations should check for instances embedded in platforms and internal tools, as well as standalone deployments. It also advises reviewing gateway access and permissions, and rotating stored provider keys if there are signs of unauthorized access. The researchers say AI gateways should be included in security assessments alongside web applications and cloud infrastructure.